Separation of man-made and natural patterns in high-altitude imagery of agricultural areas

A nonstationary linear digital filter is designed and implemented which extracts the natural features from high-altitude imagery of agricultural areas. Essentially, from an original image a new image is created which displays information related to soil properties, drainage patterns, crop disease, and other natural phenomena, and contains no information about crop type or row spacing. A model is developed to express the recorded brightness in a narrow-band image in terms of man-made and natural contributions and which describes statistically the spatial properties of each. The form of the minimum mean-square error linear filter for estimation of the natural component of the scene is derived and a suboptimal filter is implemented. Nonstationarity of the two-dimensional random processes contained in the model requires a unique technique for deriving the optimum filter. Finally, the filter depends on knowledge of field boundaries. An algorithm for boundary location is proposed, discussed, and implemented.